% Encoding: UTF-8
@COMMENT{BibTeX export based on data in FAU CRIS: https://cris.fau.de/}
@COMMENT{For any questions please write to cris-support@fau.de}
@inproceedings{faucris.200223306,
abstract = {In the literature, the examination timetabling problem (ETTP) is mostly described as a post enrollment problem (PE-ETTP). As such, it is known at optimization time how many students will take an exam and consequently how big a room is needed and which exams should not be held at the same time because of overlapping student lists. To compute a timetable using this approach, students need to register for exams before the timetable is generated. A direct consequence is that at registration time students have no idea when their exams are being held. Furthermore as timetables are often released at the end of the semester, it is hard for lecturers to plan their other responsibilities accordingly. This leads to a negative reaction from both the student body and the staff holding the exams. In this paper, we describe a curriculum-based examination timetabling variant that is similar to the curriculum-based examination timetabling problem model (CB-ETTP) introduced by Cataldo et al. 2017. The aim of the model introduced in this work is to combine the positive aspects of PE-ETTP and CB-ETTP by the use of machine learning while reducing the problems of the CB-ETTP, namely the overestimation in the number of students taking an exam. We describe an approach to calculate the number of students taking an exam by using old planning data. Furthermore we give an example for integrating the knowledge from past experience as a new soft constraint. Through the addition of this new soft constraint, we get a measure for the robustness of the timetable in respect to the uncertainty in the data. Finally, we present experiments based on real world data from the University of Erlangen-Nuremberg (FAU) showing that the approach gives a good estimation for the number of students

with only slight deviations from the actual numbers.